In the current server hardware market, computing capability and storage processing capability of a single processor are quite limited. However, in some special application scenarios, both the processing capability and memory capacity of a server are highly required, such as the processing of an accounting system of a bank, or the database processing of a communication exchange center, or analytical processing of meteorological information. In order to satisfy these special applications, multiple processors are often required to be connected together to form a large-scale computing system, which involves interconnection of processors or computing systems. Usually, the multiple processors or computing systems may be connected in two methods, one is to connect the multiple processors through a special protocol, and the other is to connect multiple small computing systems to form a large computing system through a general protocol (such as a PCIE (Peripheral Component Interconnect Express, peripheral component interconnect express)).
For the computing system which adopts the general protocol, if the processor of a network node needs to access a memory or an input/output (Input/Output, abbreviated as IO) space of another network node, the network node needs to send relevant commands and parameters to the another network node through a network protocol, and the processor of the another network node executes relevant access commands instead, but efficiency; in this way, efficiency and speed of the cross-node memory access and IO space access are very low, thereby seriously affecting system performance.